I’m in the same spot as you are. As I’m also just getting started, I’m not the best person to give you advice, but here’s what I’m doing.

First, there’s a distinction between running your own model or using an external API. I’d like to look into both, just to understand better what options are out there.

For the former (running your own), Elixir does have some cool toys like Bumblebee and Nx. I’m rewatching the Bumblebee launch video where José Valim creates an example Phoenix app that uses AI. Based on this, I’ll build & deploy a prototype of my own, just to go through the whole process from A to Z.

Regarding the external AI / ML APIs, I guess it’s just like with any other API. Open the docs, see what features they offer, play around. Some projects on my list to look into:

1. OpenAI APIs - just because of the ChatGPT and Dall-E craze
2. Huggingface - lots of cool community models and datasets there, but it wasn’t super straightforward to use at first sight, will give it another shot.
3. Tensorflow - seems powerful and I kept hearing about it, without trying it out
4. Look into tools from the bigger cloud providers, like Google Cloud API / Amazon ML API
5. Bonus: for inspiration, I’m reading “Life 3.0” and really enjoying it so far. Super interesting thought exercises about future scenarios, helps to get the juices flowing.

I also tried to do one of Andre Ng’s basic AI courses, and while it’s great, I didn’t find the motivation to complete it. Right now I’m more in the “just start building and have fun” camp.

My hope is that some idea for a fun project will pop into my mind while playing with these tools, once I have that specific idea everything else should be easier.

Hope this helps!